CN106203430B - A kind of conspicuousness object detecting method based on foreground focused degree and background priori - Google Patents
A kind of conspicuousness object detecting method based on foreground focused degree and background priori Download PDFInfo
- Publication number
- CN106203430B CN106203430B CN201610531085.5A CN201610531085A CN106203430B CN 106203430 B CN106203430 B CN 106203430B CN 201610531085 A CN201610531085 A CN 201610531085A CN 106203430 B CN106203430 B CN 106203430B
- Authority
- CN
- China
- Prior art keywords
- mrow
- pixel
- msub
- background
- super
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A kind of conspicuousness object detecting method based on foreground focused degree and background priori, step is as follows:First, image preprocessing;2nd, the conspicuousness based on foreground focused degree;3rd, the conspicuousness based on background priori;4th, conspicuousness optimization fusion;This method obtains concentration class feature by way of based on Hash coding and concentration class weight, and prospect notable figure is obtained after being combined with the contrast metric based on center priori;By rejecting super-pixel too high with prospect similarity in edge seed, so as to obtain background seed, and background conspicuousness is obtained by calculating each super-pixel and the diversity factor of background seed;Finally, the cost function for including background item, prospect and smooth item is built, is obtained by minimizing cost function and optimizes final conspicuousness.The uniform highlighting foreground target of notable figure energy that the present invention is obtained, and suppress ambient noise well, natural image is widely portable to, is conducive to succeeding target detection, Target Segmentation etc. to apply, with actual application value.
Description
(1) technical field
The present invention relates to a kind of conspicuousness object detecting method based on foreground focused degree and background priori, belong to computer
Vision and digital image processing field.Have broad application prospects in the fields such as Target Segmentation, target identification.
(2) background technology
Human eye is often easier to notice entirely different attribute with surrounding environment in scene, the obvious object of difference, this
Planting the ability of automatic sensing easily can separate the interesting part in scene.The exactly this and surrounding of object
The significant difference of the environment and ability for causing human eye to note is referred to as vision significance.Theoretical, the vision according to classical conspicuousness
Attention mechanism can be divided into two classes:Top-down attention mechanism and attention mechanism from bottom to top.Top-down attention machine
System is that, by task-driven, in this mechanism, the consciousness and task of people determine the well-marked target in image is which part, root
According to task, the subjective consciousness of someone sets out, and the view of people has played very big effect in this mechanism.And bottom-up note
Meaning mechanism is that self-contained information is determined by data-driven, that is, in image.Often by the object in image with
Otherness around it determines the conspicuousness of the object.
In computer vision field, top-down vision noticing mechanism is due to based on its subjective consciousness with people, research
Get up very difficult, so present conspicuousness field is found by main research of bottom-up vision noticing mechanism.The bottom of from
The most classical in the research of upward mechanism is then the Itti models proposed in 1998, and this model is extracted in multiscale space
Brightness, color, three, direction feature as image feature, by filtering and center --- surrounding difference algorithm obtains three
Three kinds of Fusion Features are finally obtained final notable figure by the characteristic pattern of individual feature with the method for linear plus sum.Liu et al. exists
By carrying out feature extraction to well-marked target, multiscale contrast, center-periphery contrast, color space point are obtained within 2007
3 kinds of features of cloth, then be combined with conditional random field models, obtain final conspicuousness testing result.Hou et al. was in 2009
A kind of conspicuousness computation model based on residual spectrum is proposed, in Fourier's domain of variation with preimage information and the difference of its redundancy
Obtain composing residual risk, spatial domain is changed to residual risk contravariant and obtains its Saliency maps.Jiang et al. is proposed for 2013
UFO models, conspicuousness is weighed by the uniqueness of combining target, centrality and like physical property characteristic.Zhu et al. is by calculating area
Length and the ratio of region area that domain profile is connected with border, have obtained the contour connection feature in region, and basic herein
On calculate the contrast based on background weight, the conspicuousness in each region is estimated finally by the method for optimization.
Conventional conspicuousness model is generally only from target, or only from background.The present invention combines foreground target
The advantage of feature and background priori, it is proposed that the computational methods and background initial point selection method of a kind of foreground focused degree, and profit
Prospect is merged with background with the mode of optimization, prospect is fully highlighted and inhibits background.
(3) content of the invention
(1) purpose of the present invention
In order to make up the deficiency of conventional method, there is provided one from prospect concentration class priori and background priori by the present invention
Plant the conspicuousness object detection method based on foreground focused degree and background priori.
Concentration class priori in the present invention combines concentration class and center priori, and background priori is gone out from the border of image
Hair.Because by substantial amounts of image viewing it can be found that aggregation is compared in distribution of the conspicuousness target in entire image, but
Background is then distributed relatively broad, is often distributed among entire image, and according to this discovery, the present invention constructs concentration class priori.Separately
Outside, it is accustomed to according to photography, target, which is normally located in image, leans on paracentral position, and many in existing method is all by image
Center is as center priori, but this method is easy to mistake occur, is based on to solve to employ in this problem, the present invention
The center priori of convex closure, can choose more reliable center priori according to image adaptive.Seen again by substantial amounts of image
Examine it can be found that being typically generally background close to the part of image boundary, then existing many methods then select the border of image
It is used as background priori.But exist a part of conspicuousness target is included in some situations, the border of image in practice, in order to tackle
This situation, the present invention proposes a kind of system of selection of background seed point, so as to provide more accurately background priori.
(2) technical scheme
A kind of conspicuousness object detecting method based on foreground focused degree and background priori of the present invention, its specific method step
It is rapid as follows:
Step one:Image preprocessing;, first, will by the gauss hybrid models for building input picture for subsequent step
Input picture is divided into multilayer, and obtains using hash conversion the binary code of each layer;Furthermore, will be inputted and schemed by super-pixel segmentation
As being divided into, many colors are similar, protect the super-pixel on border, and calculate the mean place and average color of each super-pixel;Carry in addition
The convex closure that well-marked target is included in input picture is taken, center priori is used as using convex closure center;
Wherein, " binary code that each layer is obtained using hash conversion " described in step one, its practice is as follows:First
The gauss hybrid models of input picture are built, a kind of color is represented with each composition correspondence of gauss hybrid models, then will can input
The color of image is divided into 6 classes, while obtaining the probability that each pixel belongs to all kinds of.The probability that pixel belongs to each layer can use image
To represent, then decomposed relative to by input picture for 6 parts, i.e., 6 layers of gray level image of degree of membership are represented with gray value;Then
This 6 width image is downsampled to the image that size is 8 × 8, its gray average is calculated, gray value is more than to the mark of average pixel
1 is designated as, is otherwise 0, so as to obtain corresponding 64 binary codes of every tomographic image;
Step 2:Conspicuousness based on foreground focused degree;First using the similarity degree between each layer binary code as similar
Property is estimated, and each layer of the gauss hybrid models of input picture is classified, then by calculating all kinds of aggregations based on center priori
Degree merges to progress as weight and obtains concentration class feature;The global contrast that each super-pixel combines central priori is calculated again, is obtained
To contrast metric.Finally concentration class feature is multiplied with contrast metric, foreground focused degree notable figure is used as;
Wherein, " each layer of the gauss hybrid models of input picture is classified " described in step 2, its practice is such as
Under:Inverse first using the Euclidean distance between the corresponding binary code of each tomographic image of gauss hybrid models is surveyed as similarity
Degree, is gathered this 6 tomographic image for 3 classes using Alex Rodriguez clustering method, prospect respectively in representative image, background and
Dash area, then the probability that each pixel belongs to K classes in this three class is:
Wherein p (k | Ix) it is pixel IxBelong to the probability of k-th of composition of gauss hybrid models, and this k-th one-tenth belongs to
K classes, equivalent to several tomographic images for belonging to K classes are added and.
Wherein, " it is used as weight to entering by calculating all kinds of concentration class based on center priori again described in step 2
Row fusion obtains concentration class feature ", the process that it is calculated is as follows:The three class images for obtaining classification by weight of concentration class add
With obtain concentration class characteristic pattern:
Comp (K) is the corresponding concentration class of K class images:
Step 3:Conspicuousness based on background priori;The super-pixel being connected with image boundary is obtained first as background kind
Son;Then, to the prospect notable figure binaryzation obtained in step 2, it regard the super-pixel for being marked as 1 as foreground seeds point, meter
Calculate the similarity degree of other super-pixel and foreground seeds, and threshold value;Will be big with foreground seeds similarity in the super-pixel of border
Rejected in the part super-pixel of threshold value from background seed, then obtain final background subset;Finally, by calculating each super-pixel
With the contrast of background seed, so as to obtain background conspicuousness;
Wherein, " similarity degree for calculating other super-pixel and foreground seeds " described in step 3, its computational methods
It is as follows:
FS represents foreground seeds point set.
Wherein, " part for being more than threshold value with foreground seeds similarity in the super-pixel of border being surpassed described in step 3
Pixel is rejected from background seed ", the process that it is rejected is as follows:
The threshold value T of similarity is determined by OSTU algorithms, threshold value T will be more than with foreground seeds similarity in the super-pixel of border
Part super-pixel rejected from background seed, then obtain final background subset BS;Finally, with each super-pixel and background seed
Contrast as to background conspicuousness:
Step 4:Conspicuousness optimization fusion;Fusion problem is considered as optimization problem, one is built and includes prospect, background
The cost function of item and smooth item, prospect background is combined together, final notable figure is obtained by minimizing cost function;
In described step four, a cost function is built first, prospect background is combined together:
Foreground represents prospect, and Background represents background item, and Smoothness is smooth item;Wherein S (i)
For the final conspicuousness average of i-th of super-pixel, final notable figure is obtained by minimizing cost function;α is balance prospect
Conspicuousness and weight of the background conspicuousness to final conspicuousness influence power size, λ is the weight that the smooth item of regulation acts on size, i.e.,
Adjust the smoothness of final conspicuousness.
Finally by cost function is minimized, final conspicuousness S is obtained;
By above step, this detection method combines display foreground concentration class and background priori, before can preferably protruding
Scape and suppression background, then can relatively accurately detect image object, for other image processing fields such as Target Segmentation, mesh
Mark tracking and target retrieval etc. have actual application value.
(3) compared with prior art, advantages of the present invention:
First, the present invention is used as center priori using convex closure center, it is proposed that a kind of meter of the concentration class feature at relative center
Calculation method, and it is combined with the global contrast based on center priori, more complete obvious object can be obtained, and fill
Divide the conspicuousness for the prospect that highlighted.
Secondly, the present invention proposes a kind of background seed point selection algorithm based on prospect, it is to avoid part is located at into side
The prospect on boundary is falsely dropped as background seed, so as to improve the accuracy of background priori.Conspicuousness based on background priori, which is calculated, to be had
The background parts inhibited in notable figure of effect.
Finally, prospect is considered as optimization problem processing by the present invention with merging for background conspicuousness, by building cost function
With reference to prospect and background, the advantage of prospect notable figure and background notable figure is taken full advantage of, and notable figure is seamlessly transitted, is filled
While dividing prominent prospect, background is also inhibited well.
(4) illustrate
Fig. 1 is the FB(flow block) of detection method of the present invention.
(5) embodiment
Embodiments of the present invention are made further by technical scheme for a better understanding of the present invention below in conjunction with accompanying drawing
Description.
The FB(flow block) of the present invention is as shown in figure 1, a kind of conspicuousness based on foreground focused degree and background priori of the present invention
Object detecting method, its specific implementation step is as follows:
Step one:Image preprocessing
First, the gauss hybrid models of input picture are built, a kind of face is represented with each composition correspondence of gauss hybrid models
Color, then can be divided into 6 classes by the color of input picture, while obtaining the probability that each pixel belongs to kth class color:
{ωk,μk,∑kFor the parameter of gauss hybrid models, the probability that pixel belongs to each layer can be represented with image, then
Decomposed relative to by input picture for 6 parts, i.e., 6 layers of gray level image of degree of membership are represented with gray value.Then by this 6 width figure
As being downsampled to the image that size is 8 × 8, its gray average is calculated, is by the mark that gray value is more than average pixel, it is no
It is then 0, then the binary code of one 64 is all can obtain per tomographic image.
Then, using SLIC algorithms, input picture is too segmented into M=200 super-pixel.And calculate the position of each super-pixel
Put μiWith color average ci:
Wherein IcFor the pixel I belonged toxColor vector, IμFor corresponding space coordinate vector, qiSuper-pixel block PiMiddle bag
The number of pixels contained.
Finally, Harris Corner Detections, the Harris angle point energy of calculating input image are carried out to the coloured image of input
Function obtains energy diagram, chooses the maximum several points of energy value in energy diagram, and rejects the point of near image boundaries, obtains calibrated
All significant points, are surrounded and represent marking area by true significant point with a convex closure, and first using convex closure center as center
Test.
Step 2:Conspicuousness based on foreground focused degree
First, using the reciprocal as similar of the Euclidean distance between the corresponding binary code of each tomographic image of gauss hybrid models
Degree is estimated, and is gathered this 6 tomographic image for 3 classes, prospect respectively in representative image, the back of the body using Alex Rodriguez clustering method
Scape and dash area.The probability that then each pixel belongs to K classes in this three class is:
Wherein p (k | Ix) it is pixel IxBelong to the probability of k-th of composition of gauss hybrid models, and this k-th one-tenth belongs to
K classes, equivalent to several tomographic images for belonging to K classes are added and.This three classes image is added and assembled by weight of concentration class again
Spend feature:
Comp (K) is the corresponding concentration class of K class images, and specific formula is:
X is pixel IxCoordinate position, μ is the coordinate position of picture centre.
Then, it is with reference to the calculation formula of the global contrast of center priori:
ciRepresent super-pixel i color average, μiRepresent super-pixel i position average.σpFor adjustment color and locus
The weight of influence power, σcIt is then the weight of control centre's priori influence power.
Finally, concentration class feature and contrast metric are combined in the form of multiplication and obtain final prospect conspicuousness:
Sfg(i)=SC(i)·SU(i) (7)
SC(i) super-pixel i average aggregate degree characteristic value is represented.
Step 3:Conspicuousness based on background priori
The super-pixel being connected with image boundary is obtained first as background seed.Then, the prospect to being obtained in step 2
Notable figure binaryzation, will be marked as 1 super-pixel as foreground seeds point, calculates other super-pixel similar to foreground seeds
Degree, specific formula is:
Wherein FS represents foreground seeds point set.
The threshold value T of similarity is determined by OSTU algorithms, threshold value T will be more than with foreground seeds similarity in the super-pixel of border
Part super-pixel rejected from background seed, then obtain final background subset BS.Finally, each super-pixel and background kind are calculated
The contrast of son, so as to obtain background conspicuousness, specific formula is:
Step 4:Conspicuousness optimization fusion
A cost function for including prospect, background item and smooth item is built, prospect background is combined together, specifically
Formula is:
Foreground represents prospect, and Background represents background item, and Smoothness is smooth item.Wherein S (i)
For the final conspicuousness average of i-th of super-pixel, final notable figure is obtained by minimizing cost function.α is balance prospect
Conspicuousness and weight of the background conspicuousness to final conspicuousness influence power size, λ is the weight that the smooth item of regulation acts on size, i.e.,
Adjust the smoothness of final conspicuousness.
Finally by cost function is minimized, final conspicuousness S is obtained.
Claims (5)
1. a kind of conspicuousness object detecting method based on foreground focused degree and background priori, it is characterised in that:Its specific method
Step is as follows:
Step one:Image preprocessing;First, it is divided into multilayer by building the gauss hybrid models of input picture by input picture,
And the binary code of each layer is obtained using hash conversion;Furthermore, input picture is divided into by many colors by super-pixel segmentation
It is similar, the super-pixel on border is protected, and calculate the mean place and average color of each super-pixel;Extract and included in input picture in addition
The convex closure of well-marked target, center priori is used as using convex closure center;
Step 2:Conspicuousness based on foreground focused degree;Surveyed first using the similarity degree between each layer binary code as similitude
Degree, each layer of the gauss hybrid models of input picture is classified, then is made by calculating all kinds of concentration class based on center priori
Merged for weight, obtain concentration class feature;The global contrast that each super-pixel combines central priori is calculated again, is contrasted
Spend feature;Finally concentration class feature is multiplied with contrast metric, foreground focused degree notable figure is used as;
Step 3:Conspicuousness based on background priori;The super-pixel being connected with image boundary is obtained first as background seed;So
Afterwards, to the prospect notable figure binaryzation obtained in step 2,1 super-pixel will be marked as foreground seeds point, it is calculated
The similarity degree of his super-pixel and foreground seeds, and threshold value;Threshold will be more than with foreground seeds similarity in the super-pixel of border
The part super-pixel of value is rejected from background seed, then obtains final background subset;Finally, by calculating each super-pixel and the back of the body
The contrast of scape seed, so as to obtain background conspicuousness;
Step 4:Conspicuousness optimization fusion;Fusion problem is considered as optimization problem, build one comprising prospect, background item and
The cost function of smooth item, prospect background is combined together, final notable figure is obtained by minimizing cost function;
In described step four, a cost function is built first, prospect background is combined together:
Foreground represents prospect, and Background represents background item, and Smoothness is smooth item;Wherein S (i) is the
The final conspicuousness average of i super-pixel, final notable figure is obtained by minimizing cost function;α is that balance prospect is notable
Property with background conspicuousness to the weight of final conspicuousness influence power size, λ is the weight that the smooth item of regulation acts on size, that is, is adjusted
The smoothness of final conspicuousness;
Finally by cost function is minimized, final conspicuousness S is obtained;
Sfg(i) it is the corresponding prospect saliency value of super-pixel i;Sbg(i) it is the corresponding background conspicuousnesses of super-pixel i;
By above step, this detection method combines display foreground concentration class and background priori, can preferably protrude prospect with
Suppress background, then relatively accurately detect image object, for other image processing fields such as Target Segmentation, target following and
Target retrieval has actual application value;
Wherein, described in step 2 " be used as weight by calculating all kinds of concentration class based on center priori again to be merged,
Obtain concentration class feature ", the process that it is calculated is as follows:The three class images for obtaining classification by weight of concentration class add and obtained
Concentration class characteristic pattern:
<mrow>
<msub>
<mi>S</mi>
<mi>C</mi>
</msub>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mi>K</mi>
</munder>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>K</mi>
<mo>|</mo>
<msub>
<mi>I</mi>
<mi>x</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>*</mo>
<mi>C</mi>
<mi>o</mi>
<mi>m</mi>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>K</mi>
<mo>)</mo>
</mrow>
</mrow>
Comp (K) is the corresponding concentration class of K class images:
<mrow>
<mi>C</mi>
<mi>o</mi>
<mi>m</mi>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>K</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msup>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<msub>
<mi>I</mi>
<mi>x</mi>
</msub>
</msub>
<mo>|</mo>
<mo>|</mo>
<mi>x</mi>
<mo>-</mo>
<mi>&mu;</mi>
<mo>|</mo>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
<mo>&CenterDot;</mo>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>K</mi>
<mo>|</mo>
<msub>
<mi>I</mi>
<mi>x</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mo>&Sigma;</mo>
<msub>
<mi>I</mi>
<mi>x</mi>
</msub>
</msub>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>K</mi>
<mo>|</mo>
<msub>
<mi>I</mi>
<mi>x</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mo>;</mo>
</mrow>
1
Wherein, x is pixel IxCoordinate position, μ is the coordinate position of picture centre.
2. a kind of conspicuousness object detecting method based on foreground focused degree and background priori according to claim 1, its
It is characterised by:" binary code that each layer is obtained using hash conversion " described in step one, its way is as follows:Build first
The gauss hybrid models of input picture, represent a kind of color, then by input picture with each composition correspondence of gauss hybrid models
Color is divided into 6 classes, while obtaining the probability that each pixel belongs to all kinds of;The probability that pixel belongs to each layer can represent with image,
Then decomposed relative to by input picture for 6 parts, i.e., 6 layers of gray level image of degree of membership are represented with gray value;Then by this 6 width
Image is downsampled to the image that size is 8 × 8, calculates its gray average, is by the mark that gray value is more than average pixel,
Otherwise it is 0, so as to obtain corresponding 64 binary codes of every tomographic image.
3. a kind of conspicuousness object detecting method based on foreground focused degree and background priori according to claim 1, its
It is characterised by:" each layer of the gauss hybrid models of input picture is classified " described in step 2, its way is as follows:It is first
First using the reciprocal as similarity measure of the Euclidean distance between the corresponding binary code of each tomographic image of gauss hybrid models, utilize
Alex Rodriguez clustering method gathers this 6 tomographic image for 3 classes, respectively prospect, background and the shadow part in representative image
Point, then the probability that each pixel belongs to K classes in this three class is:
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>K</mi>
<mo>|</mo>
<msub>
<mi>I</mi>
<mi>x</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>&Element;</mo>
<mi>K</mi>
</mrow>
</munder>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>|</mo>
<msub>
<mi>I</mi>
<mi>x</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
Wherein p (k | Ix) it is pixel IxBelong to the probability of k-th of composition of gauss hybrid models, and this k-th one-tenth belongs to K classes,
Equivalent to several tomographic images for belonging to K classes are added and.
4. a kind of conspicuousness object detecting method based on foreground focused degree and background priori according to claim 1, its
It is characterised by:" similarity degree for calculating other super-pixel and foreground seeds " described in step 3, its computational methods is as follows:
<mrow>
<mi>S</mi>
<mi>i</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mn>1</mn>
<mo>-</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>&Element;</mo>
<mi>F</mi>
<mi>S</mi>
</mrow>
</munder>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>S</mi>
<mrow>
<mi>f</mi>
<mi>g</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<msub>
<mi>S</mi>
<mrow>
<mi>f</mi>
<mi>g</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>|</mo>
<mo>|</mo>
</mrow>
FS represents foreground seeds point set, Sfg(j) it is the corresponding prospect saliency value of super-pixel j.
5. a kind of conspicuousness object detecting method based on foreground focused degree and background priori according to claim 1, its
It is characterised by:" the part super-pixel of threshold value will be more than in the super-pixel of border with foreground seeds similarity described in step 3
Rejected from background seed ", the process that it is rejected is as follows:
The threshold value T of similarity is determined by OSTU algorithms, threshold value T portion will be more than in the super-pixel of border with foreground seeds similarity
Divide super-pixel to be rejected from background seed, then obtain final background subset BS;Finally, with each super-pixel and pair of background seed
Than degree as to background conspicuousness:
<mrow>
<msub>
<mi>S</mi>
<mrow>
<mi>b</mi>
<mi>g</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>&Element;</mo>
<mi>B</mi>
<mi>S</mi>
</mrow>
</munder>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>c</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>c</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
<mo>|</mo>
<mo>&CenterDot;</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>&mu;</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>&mu;</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
<mo>|</mo>
<mo>)</mo>
</mrow>
</mrow>
Wherein, ciRepresent super-pixel i color average, μiRepresent super-pixel i position average;cjThe color for representing super-pixel j is equal
Value, μjRepresent super-pixel j position average.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610531085.5A CN106203430B (en) | 2016-07-07 | 2016-07-07 | A kind of conspicuousness object detecting method based on foreground focused degree and background priori |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610531085.5A CN106203430B (en) | 2016-07-07 | 2016-07-07 | A kind of conspicuousness object detecting method based on foreground focused degree and background priori |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106203430A CN106203430A (en) | 2016-12-07 |
CN106203430B true CN106203430B (en) | 2017-11-03 |
Family
ID=57472414
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610531085.5A Expired - Fee Related CN106203430B (en) | 2016-07-07 | 2016-07-07 | A kind of conspicuousness object detecting method based on foreground focused degree and background priori |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106203430B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107833220A (en) * | 2017-11-28 | 2018-03-23 | 河海大学常州校区 | Fabric defect detection method based on depth convolutional neural networks and vision significance |
Families Citing this family (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106846331A (en) * | 2016-12-22 | 2017-06-13 | 中国科学院文献情报中心 | Joint vision significance and figure cut the image automatic segmentation method of optimization |
CN106651853B (en) * | 2016-12-28 | 2019-10-18 | 北京工业大学 | The method for building up of 3D conspicuousness model based on priori knowledge and depth weight |
CN106780422A (en) * | 2016-12-28 | 2017-05-31 | 深圳市美好幸福生活安全系统有限公司 | A kind of notable figure fusion method based on Choquet integrations |
CN106778903A (en) * | 2017-01-09 | 2017-05-31 | 深圳市美好幸福生活安全系统有限公司 | Conspicuousness detection method based on Sugeno fuzzy integrals |
CN108537819A (en) * | 2017-03-03 | 2018-09-14 | 防城港市港口区思达电子科技有限公司 | Super-pixel moving target detecting method |
CN108537242A (en) * | 2017-03-03 | 2018-09-14 | 防城港市港口区思达电子科技有限公司 | A kind of new conspicuousness object detection method |
CN107133558B (en) * | 2017-03-13 | 2020-10-20 | 北京航空航天大学 | Infrared pedestrian significance detection method based on probability propagation |
CN107424142B (en) * | 2017-03-30 | 2020-05-19 | 上海万如科技发展有限公司 | Weld joint identification method based on image significance detection |
CN107085725B (en) * | 2017-04-21 | 2020-08-14 | 河南科技大学 | Method for clustering image areas through LLC based on self-adaptive codebook |
CN107766857B (en) * | 2017-10-17 | 2021-08-03 | 天津大学 | Visual saliency detection algorithm based on graph model construction and label propagation |
CN107886507B (en) * | 2017-11-14 | 2018-08-21 | 长春工业大学 | A kind of salient region detecting method based on image background and spatial position |
CN107992874B (en) * | 2017-12-20 | 2020-01-07 | 武汉大学 | Image salient target region extraction method and system based on iterative sparse representation |
CN108416347A (en) * | 2018-01-04 | 2018-08-17 | 天津大学 | Well-marked target detection algorithm based on boundary priori and iteration optimization |
CN108416768B (en) * | 2018-03-01 | 2021-05-25 | 南开大学 | Binary-based foreground image similarity evaluation method |
CN108537816B (en) * | 2018-04-17 | 2021-08-31 | 福州大学 | Salient object segmentation method based on superpixel and background connection prior |
CN108805139B (en) * | 2018-05-07 | 2022-02-18 | 南京理工大学 | Image similarity calculation method based on frequency domain visual saliency analysis |
CN110245659B (en) | 2019-05-21 | 2021-08-13 | 北京航空航天大学 | Image salient object segmentation method and device based on foreground and background interrelation |
CN110287802B (en) * | 2019-05-29 | 2022-08-12 | 南京邮电大学 | Human eye gaze point prediction method based on optimized image foreground and background seeds |
CN110827309B (en) * | 2019-11-12 | 2023-06-23 | 太原理工大学 | Super-pixel-based polaroid appearance defect segmentation method |
CN111144294A (en) * | 2019-12-26 | 2020-05-12 | 上海眼控科技股份有限公司 | Target identification method and device, computer equipment and readable storage medium |
CN111080748B (en) * | 2019-12-27 | 2023-06-02 | 北京工业大学 | Automatic picture synthesizing system based on Internet |
CN111815610B (en) * | 2020-07-13 | 2023-09-12 | 广东工业大学 | Lesion detection method and device for lesion image |
CN113379691B (en) * | 2021-05-31 | 2022-06-24 | 南方医科大学 | Breast lesion deep learning segmentation method based on prior guidance |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102722891A (en) * | 2012-06-12 | 2012-10-10 | 大连理工大学 | Method for detecting image significance |
CN103914834A (en) * | 2014-03-17 | 2014-07-09 | 上海交通大学 | Significant object detection method based on foreground priori and background priori |
CN103996198A (en) * | 2014-06-04 | 2014-08-20 | 天津工业大学 | Method for detecting region of interest in complicated natural environment |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9245187B1 (en) * | 2014-07-07 | 2016-01-26 | Geo Semiconductor Inc. | System and method for robust motion detection |
US9454712B2 (en) * | 2014-10-08 | 2016-09-27 | Adobe Systems Incorporated | Saliency map computation |
-
2016
- 2016-07-07 CN CN201610531085.5A patent/CN106203430B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102722891A (en) * | 2012-06-12 | 2012-10-10 | 大连理工大学 | Method for detecting image significance |
CN103914834A (en) * | 2014-03-17 | 2014-07-09 | 上海交通大学 | Significant object detection method based on foreground priori and background priori |
CN103996198A (en) * | 2014-06-04 | 2014-08-20 | 天津工业大学 | Method for detecting region of interest in complicated natural environment |
Non-Patent Citations (2)
Title |
---|
Contrast and Distribution based Saliency Detection in Infrared Images;Lu Li等;《2015 IEEE 17th International Workshop on Multimedia Signal Processing (MMSP)》;20151021;1-6 * |
Saliency Optimization from Robust Background Detection;Wangjiang Zhu等;《2014 IEEE Conference on Computer Vision and Pattern Recognition》;20140628;2814-2821 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107833220A (en) * | 2017-11-28 | 2018-03-23 | 河海大学常州校区 | Fabric defect detection method based on depth convolutional neural networks and vision significance |
CN107833220B (en) * | 2017-11-28 | 2021-06-11 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
Also Published As
Publication number | Publication date |
---|---|
CN106203430A (en) | 2016-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106203430B (en) | A kind of conspicuousness object detecting method based on foreground focused degree and background priori | |
CN108549891B (en) | Multi-scale diffusion well-marked target detection method based on background Yu target priori | |
CN106886023B (en) | A kind of Radar Echo Extrapolation method based on dynamic convolutional neural networks | |
CN110909690B (en) | Method for detecting occluded face image based on region generation | |
CN111415316B (en) | Defect data synthesis method based on generation countermeasure network | |
Amuti et al. | Analysis of land cover change and its driving forces in a desert oasis landscape of Xinjiang, northwest China | |
CN105869178B (en) | A kind of complex target dynamic scene non-formaldehyde finishing method based on the convex optimization of Multiscale combination feature | |
CN106296695A (en) | Adaptive threshold natural target image based on significance segmentation extraction algorithm | |
CN110020606A (en) | A kind of crowd density estimation method based on multiple dimensioned convolutional neural networks | |
CN107256225A (en) | A kind of temperature drawing generating method and device based on video analysis | |
CN108830205A (en) | Based on the multiple dimensioned perception pedestrian detection method for improving full convolutional network | |
CN108647741A (en) | A kind of image classification method and system based on transfer learning | |
CN110532970A (en) | Age-sex's property analysis method, system, equipment and the medium of face 2D image | |
CN107886507B (en) | A kind of salient region detecting method based on image background and spatial position | |
CN108961675A (en) | Fall detection method based on convolutional neural networks | |
CN107564022A (en) | Saliency detection method based on Bayesian Fusion | |
CN107330875A (en) | Based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images | |
CN107730515A (en) | Panoramic picture conspicuousness detection method with eye movement model is increased based on region | |
CN104794693B (en) | A kind of portrait optimization method of face key area automatic detection masking-out | |
CN104282008B (en) | The method and apparatus that Texture Segmentation is carried out to image | |
CN104952070B (en) | A kind of corn field remote sensing image segmentation method of class rectangle guiding | |
CN108664838A (en) | Based on the monitoring scene pedestrian detection method end to end for improving RPN depth networks | |
CN112766184B (en) | Remote sensing target detection method based on multi-level feature selection convolutional neural network | |
CN109902613A (en) | A kind of human body feature extraction method based on transfer learning and image enhancement | |
CN110084284A (en) | Target detection and secondary classification algorithm and device based on region convolutional neural networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171103 Termination date: 20200707 |
|
CF01 | Termination of patent right due to non-payment of annual fee |